ES2283853T3 - TURBO-DECODING WITH ITERATIVE ESTIMATION OF CHANNEL PARAMETERS. - Google Patents

Abstract

Method and decoding device for decoding a convolutionally coded input data signal y. The input data signal is multiplied with a scaling factor Lc(8) and then demultiplexed (6). The demultiplexed input data signal LcS is then turbo decoded (5) in order to obtain decoder output likelihood ratio data Λ. The scaling factor Lc is updated (7) for a next iteration in dependence on a combination of a posteriori likelihood data based on turbo decoded output data Λ and a priori likelihood data based on the demultiplexed signal LcS.

Description

Turbo decoding with iterative estimation of channel parameters.

Field of the Invention

The present invention relates to a method for decoding a signal and input data, encoded by convolution, comprising the multiplication of the input data signal by a scale factor L c, to demultiplex the multiplied signal L and of input data in three signals that are related to systematic bits and with parity bits, a demultiplexed input data signal L c S that is associated with the systematic bits, for example, in signals of parity and a systematic signal, and decoding by means of turbo decoding the signal L c S of demultiplexed input data to obtain output data A decoded by turbo decoding. In a further aspect, the present invention relates to a decoder device as defined in the preamble of claim 5.

Prior art

Such a method to decode data and a decoder device are known from the patent US-B-6,574,291 which describes a turbo code decoder with iterative estimation of channel parameters.

The US Patent Publication US-B-6,393,076 describes a method to decode turbo codes using scale variation of data. The input data encoded by convolution are Decoded almost ideally. A part of the data of entry is temporarily stored in memory, after which an average of the data of the data part of entry. Then, a mean square value of the part is calculated using the average. A scale variation factor is derived from mean square value, whose scale variation factor is use to scale the part of the input data before the turbo decoding operation.

These described methods have the disadvantage that the calculation of the scale variation factor is not based on a posteriori probability information, which leads to an additional loss. Especially in mobile applications that use these types of coding, each additional loss has a negative effect on the system's behavior.

The publication of El-Gamal, H., High capacity Synchronous FH / SSMA networks with turbo decoding, (High capacity synchronous FH / SSMA networks with decoding turbo), IEEE Intern. Symp on Spread Spectrum Techniques, 1998, P. 973-977, provides a similar method for convert input data to scale. The method consists of calculate the variation of the noise, which is applied subsequently to form logarithmic probability values. This method has the drawback of the signal amplitude normalized to the unity. This is an invalid simplification for practical equipment of reception and, therefore, negatively affects the system behavior

The article by MC Valenti et al . "Iterative channel estimation and decoding of pilot symbol assisted turbo codes over flat-fading channels", IEEE Journal on selected areas in communications, September 2001, IEEE, USA, vol. 19, no. 9, pages 1,697-1,705, describes a method for coherently detecting and decoding binary signals modulated by displacement and encoded by turbo coding, offset key, transmitted by frequency flat fade channels. Consider the use of decision feedback using hardware and decision feedback using software.

British patent application GB-A-2 360 425 describes an estimate of channel status information for turbo code decoders. In this document, only the use of a maximum- to-posterior probability decoding algorithm (MAP) is described.

Summary of the Invention

The present invention attempts to provide an improved decoding scheme for use in transceivers that use a maximum a posteriori decoder (MAP), or related techniques such as MAP logarithm (LOGMAP).

According to the present invention, a method according to the preamble defined above is provided which has the characteristics of the characterizing part of claim 1. By combining this probability data a posteriori and a priori , it is possible to improve the behavior of the decoding method without requiring Many additional hardware or software resources. In one embodiment of the present invention, the scale variation factor is updated according to:

\hskip0,3cm

en donde \hat{L}_{c} es el factor de variación de escala actualizado.\ hat {L} _ {c} = \ frac {2} {\ hat {c} \ cdot \ hat {\ sigma} _ {n '} {} ^ {2}} \ cdot L_ {c},

 \ hskip0,3cm 

where \ hat {L} c is the updated scale variation factor.

This makes the present method accurate and fast, so that it ensures the benefit (simulations show that the loss with respect to the theoretical optimum is of around 0.03 dB), and is applicable in fading channels fast or on time varying channels. It has been shown that you can get around 0.1 to 0.2 dB of sensitivity enhancement in turbo codes, compared to the methods of the technique previous.

In a further embodiment, the scale variation factor L c is initialized either as a fixed value, as a result of an initial number of iterations using a known algorithm, as the result of filtering on subsequent iterations and blocks. coding, or as the result of SNR / SIR estimates of the input data signal. Initiation can, therefore, be carried out using very simple solutions or more complex but well known solutions.

In a further embodiment of the present invention, the method further comprises calculating the variation of the scale variation factor in subsequent iterations and, when the variation, after a predetermined number of iterations, is above a predetermined threshold value, resort to a different method of calculating the scale variation factor and / or a a turbo decoding method. The different method for calculate the scale variation factor can be for example use a fixed scale variation factor, a method of maximum logarithm or a SOVA method (Viterbi Output Algorithm of software). In this way, supervision can be carried out. simple of the divergence of the iterative method, using algorithms known as reserve measure.

In a further aspect of the present invention, a decoder device is provided according to the claim 5. Additional embodiments of the present decoder device are described in the claims Dependents According to this, the present decoder device provides improved behavior on devices of the prior art Since the improvement in sensitivity requires, Only, a small hardware cost, the improvement of Sensitivity is pure benefit. The important aspects of the system that influence the sensitivity are the signal strength, for example, range / coverage / battery life, magnitude of noise, and interference, that is, air interface capacity. Therefore, the advantage of the present invention can be translated increasing coverage (<1% more range), longer duration of the battery (2 to 4% less transmission power), less strict noise magnitude requirements (0.2 dB less), or increase in possible users in the air interface (2 to 4% more). This produces greater cellular coverage and the use of less power by mobile, which, in turn, produces less interference and, by Therefore, more capacity.

In yet another aspect, the present invention provides a computer program product, which comprises a computer executable code that, when loaded in a treatment system, provides the treatment system with ability to execute the present method. System treatment can comprise a microprocessor and equipment peripheral, a digital signal processor or a combination of both to execute the present method.

Brief description of the drawings

The present invention will be described with more detail below, by means of several illustrative embodiments, referring to the attached drawing, in which

Figure 1 shows a block diagram of an embodiment of the decoding scheme according to the present invention.

Detailed description of illustrative embodiments

The present application can be applied, advantageously, in the decoding of systematic, direct error correction codes, such as parallel concatenation convolutional turbo codes found in third generation mobile telecommunication systems (3GPP). The present invention can be used in transceivers that use a maximum a posteriori decoder (MAP), or related techniques, such as the Logarithm MAP.

The decoding of iterative MAP and of MAP logarithms are optimal asymptotically. The algorithms of decoding less than optimal, such as the algorithm of Viterbi software output (SOVA) or LOGMAP (logarithm maximum) approximate, present a more practical way of execution simple, up to about 0.5 dB less in relation to the MAP and MAP Logarithm behavior.

The problem to reach the full extent of the MAP behavior is the requirement that the input data are defined as odds ratio (logarithmic). With the present invention is possible to scale the data of decoder input on a time-varying channel, with fading, quickly and accurately, to form values of probability (logarithmic).

The present method uses a posteriori probability data to form the probability information of the decoded bits. This is combined with a priori probability data from the decoder, to determine, optimally, the mean of the desired signal and the noise variance. These are used to iteratively scale the information a priori .

The scale variation factor must be initialized The invention provides several different methods for do this:

- use the scale variation factor of previous block of coding (filtered or not);

- run the first iteration (or iterations) with SOVA or with maximum logarithm;

- display any method as a setting for initial value; or

- any combination of these.

Figure 1 shows a block diagram of the iterative algorithm 10 for scaling to decode data encoded by convolution. The symbols and received are multiplied by a scale variation factor L c using a multiplier 8 before they are demultiplexed using a demultiplexer 6 in systematic bits s i and par parity bits, even 2, which are introduced into a turbo decoder 5. After an iteration of decoder 5, an adaptive multiplier circuit 7 estimates the amplitude of the signal and the noise variance, based on the \ Lambda_ { i} logarithmic probability of the decoder output. From these estimates, a new scale variation factor L c is calculated for the next iteration.

The decoder 5 turbo comprises, as is shown in figure 1, interleaving elements 13,14, a collation eliminator element 15, and decoders 11, 12. In more detail, the turbo decoder 5 comprises sections first and second 11, 12 of software input decoder, software output (SISO), which receive data from the demultiplexer 6. The operation of the turbo decoder 5 is known by the expert in the art (see, for example, the document US-B-6,393,076) and is not necessary No other explanation in this description.

In what follows, the adaptive gearbox 7 action. Calculations made by adaptive gearbox 7 circuit is describe as algorithms, calculations and the like.

For experts in the field, it is of course algorithms and calculations, as performed by the various blocks and elements of the decoder algorithm 10, it they can execute in practice using software resources, from hardware or a combination of both, such as electronics analog, logic circuits, signal processors, etc.

The signal amplitude is computed in block 20 from the relations \ _ {i} Lambda log likelihood resulting from the most recent iteration of the turbo decoder 5, and the systematic bits as follows:

100

where N is the number of bits in an encoding block and s i is the systematic bit of order i. \ hat {c} is the estimation of the amplitude of the systematic bits L c \ s converted to scale; if \ hat {E} s is the estimated symbol energy of the signal component of the input sequence s i, then \ sqrt {\ hat {E} s} is the estimated amplitude of this component of the signal; then we found that

101

In block 21 the estimated amplitude \ hat {c} is used to normalize the systematic bits s ' i giving:

102

From the \ Lambda_ {i} relationships of Logarithmic probability, can be calculated in block 22 a estimation of the logical bit probabilities; the probability of that the order bit i is 0 is estimated by

103

and the probability that the bit is one through

104

The nonlinear relationship between \ Lambda_ {i} and the logical probabilities of bits Pi (0) and Pi (l) are can calculate or compute based on a table to query.

These probabilities, together with the standardized systematic bits s 'i, can be used to calculate an estimate \ hat {\ sigma} _ {n'} 2 of the noise variance in block 23, according to

105

where K is a partial correction calculated as follows:

106

Then the relationship between \ hat {\ sigma} 2 {n '} and \ hat {E} s and \ hat {N} 0 is given by

107

When the above equations are rewritten from way

108

Y use

109

the new variation factor of estimated optimal scale can be calculated in block 24 through:

110

In short, the new variation factor of scale in the estimates of amplitude and variance, and in the Current scale variation factor can be expressed as follow:

111

This calculation is performed using the multiplier 25 and the memory retention circuit 26, after which the current scale variation factor is applied, again, to the data and input.

Since the turbo decoder 5 is iterative in itself, combining it with another iterative algorithm (that of the adaptive demultiplier circuit 7) requires careful consideration to avoid divergence or oscillation. The initialization of the scale variation factor L c at a reasonable starting value proves to be sufficient. In the absence of appropriate initialization information, the first iterations of decoding can be done with known Maximum Logarithm or SOVA techniques. These are not dependent on the scale variation factor, and can therefore be used to allow the convergence of posterior probability values. Other techniques for acquiring initialization information are filtering on subsequent iterations and coding blocks, and the estimation of SNR / SIR at the input, see, for example, the method described in Summers, TA, Wilson, SG, SNR Mismatch and Online Estimation in Turbo Decoding, IEEE Tr. On COM, vol. 46, no. 4, April 1998, p. 421-423.

Moreover, divergence can be detected by monitoring the development of {c} L scale variation factor. Normally, it should converge in a few iterations, and then just change. In the case of detected divergence, one can set the scale variation factor L c, or change to Maximum Logarithm or SOVA.

The description of the previous embodiment, set forth in the claims, is based on the type of Logarithm MAP of the turbo decoder. A person skilled in the technique will recognize that the ideas of the described method are not restricted to decoding MAP Logarithm, but may be formulated, also, for variations of the method of MAP Logarithm decoding, and for MAP decoding.

Claims (9)

1. Method for decoding a signal and input data encoded by convolution, comprising - multiply the input data signal by a scale variation factor L c; - demultiplexing the signal L_ {c} and input data multiplied into three signals that are related to systematic bits and parity bits, with a signal L c S being input data, demultiplexed, associated with the systematic bits; - performing a turbo decoding of the L c S signal of demultiplexed input data to obtain decoded output data? via turbo decoding, characterized in that the scale variation factor L c is updated for a subsequent iteration based on a combination of a posteriori probability data based on output data \ Lambda decoded by turbo decoding and a priori probability data based on the demultiplexed signal L c S , using an estimate of the mean value of the amplitude \ hat {c} of the signal and an estimate of the noise variation \ hat {\ sigma} 2 {n '} in which the estimate of the average value of the signal amplitude is equal to 112 where N is the number of bits in an encoding block of the input data signal, s i is the systematic bit of order i, and \ Lambda_ {i} is the logarithmic probability ratio resulting from the iteration most recent turbo decoder, and in which the estimate of the variance of noise \ hat {\ sigma} 2 {{n '} is equal to 113 the probability that the bit of order i be 0 is estimated through 114 and the probability that the bit is one through 115 the standardized systematic bits s ' i are calculated by 116 and where K is a partial correction calculated by 117

         \ newpage
      

2. Method according to claim 1, wherein the scale variation factor is updated according to
\ hat {L} _ {c} = \ frac {2} {\ hat {c} \ cdot \ hat {\ sigma} ^ {2} _ {n '}} \ cdot L_ {c},

 \ hskip0,3cm 

where \ hat {L} c is the updated scale variation factor. 3. Method according to one of the preceding claims, wherein the scale variation factor L c is initialized either as a fixed value, as a result of an initial number of iterations using a known algorithm, as the result of filtering on subsequent iterations and code blocks, or as the result of the SNR / SIR estimation of the signal and input data. 4. Method according to one of the claims precedents, which also includes calculating the variation of scale change factor in subsequent iterations and, when the variation, after a predetermined number of iterations, is above a predetermined threshold value, resort to a different method of calculating the scale variation factor and / or turbo decoding method. 5. Decoder device for decoding a signal and input data encoded by convolution, comprising - a multiplication element (8) to multiply a signal and input data received by a scale variation factor L c; - a demultiplexer (6) for demultiplexing the signal L c and input data multiplied into three signals that are related to systematic bits and parity bits, a signal L c S being input data demultiplexed associated with systematic bits; - a turbo decoder (5) to decode the demultiplexed input data signal to obtain data Output lambda decoded by decoding Turbo, characterized in that the decoding device (10) further comprises an adaptive scale variation element (7) which is arranged to update the scale variation factor L c for a subsequent iteration based on a combination of data of posteriori probability based data \ Lambda output decoded by turbo decoding and data prior probability based on the signal L {c} S demultiplexed, using an estimate of the mean value \ hat {c} signal amplitude and an estimate of the variation \ hat {\ sigma} 2 {n '} of noise in which the estimate of the average value of the signal amplitude is equal to

         \ vskip1.000000 \ baselineskip
      

118 where N is the number of bits of an encoding block of the input data signal, s i is the systematic bit of order i, and \ Lambda_ {i} is the logarithmic probability ratio resulting from the iteration most recent turbo decoder, and in which the variance estimate \ hat {\ sigma} 2 {n '} of noise is equal to

         \ vskip1.000000 \ baselineskip
      

119 the probability that the bit of order i be 0 is estimated through

         \ vskip1.000000 \ baselineskip
      

120 and the probability that the bit is one through 121

         \ newpage
      

the standardized systematic bits s ' i are calculated according to 122 and where K is a partial correction calculated as 123 6. Decoding device according to claim 5, whose adaptive scale variation element (7) is further arranged to update the scale variation factor according to \ hat {L} _ {c} = \ frac {2} {\ hat {c} \ cdot \ hat {\ sigma} ^ {2} _ {n '}} \ cdot L_ {c},

 \ hskip0,3cm 

where \ hat {L} c is the updated scale variation factor. 7. Decoding device according to one of claims 5 and 6, wherein the decoding device is further arranged to initialize the scale variation factor L c as well as a fixed value, as a result of an initial number of iterations that use a known algorithm, such as the result of filtering over subsequent iterations and coding blocks, or as the result of the SNR / SIR estimation of the signal and input data. 8. Decoder device according to one of the claims 5-7, wherein the device decoder is also arranged to calculate the variation of the scale change factor in subsequent iterations and, when the variation, after a predetermined number of iterations, it is above a predetermined threshold value, resort to a different method of calculating the variation factor of scale and / or turbo decoding method. 9. Computer program product, which comprises a computer executable code that, when loaded  in a treatment system, it provides the treatment system the ability to execute the method according to one of the claims 1 to 4.